MSC-Thesis-Wireless-Video-over-HIPERLAN2-JustinJohnstone

DEPARTMENT OF ELECTRICAL AND ELECTRONIC
ENGINEERING
UNIVERSITY OF BRISTOL
________________________________________
“WIRELESS VIDEO OVER HIPERLAN/2”
A THESIS SUBMITTED TO THE UNIVERSITY OF BRISTOL IN ACCORDANCE WITH THE
REQUIREMENTS OF THE DEGREE OF M.SC. IN COMMUNICATIONS SYSTEMS AND
SIGNAL PROCESSING IN THE FACULTY OF ENGINEERING.
________________________________________
JUSTIN JOHNSTONE
OCTOBER 2001

i
ABSTRACT
In this report, techniques are investigated to optimise the performance of scalable modes of
video coding over a HiperLAN/2 wireless LAN.
This report commences with a review of video coding theory and overviews of the H.263+
video coding standard and HiperLAN/2. A review of the nature of errors in wireless
environments and how these errors impact the performance of video coding is then presented.
Existing packet video techniques are reviewed, and these lead to the proposal of an approach
to convey scaled video over HiperLAN/2. The approach is based on the Unequal Packet-Loss
Protection (UPP) technique, and it prioritises layered video data into four priority streams,
and associates each of these priority streams with an independent HiperLAN/2 connection.
Connections are configured to provide increased amounts of protection for higher priority
video data.
A simulation system is integrated to allow benefits of this UPP approach to be assessed and
compared against an Equal Packet-Loss Protection (EPP) scheme, where all video data shares
the same HiperLAN/2 connection. Tests conclude that the UPP approach does improve
recovered video performance. The case is argued that the UPP approach increases potential
capacity offered by HiperLAN/2 to provide multiple simultaneous video services. This
combination of improved performance and increased capacity allows the UPP approach to be
endorsed.
Testing highlighted the video decoder software’s intolerance to data loss or corruption. A
number of options to treat errored packets and video frames, prior to passing them to the
decoder, are implemented and assessed. Two options which improve recovered video
performance are recommended within the context of this project. It is also recommended that
the decoder program is modified to improve it’s robustness to data loss/corruption.
Error models within the simulation system permitted configuration of bursty errors, and tests
illustrate improved video performance under increasingly bursty error conditions. Based on
these performance differences, and the fact that bursty errors are more representative of the
nature of wireless channels; it is recommended that any future simulations include bursty
error models.

ii
ACKNOWLEDGMENTS
I would like to thank the following for their support with this project:
 David Redmill, my project supervisor at Bristol University, for his patient guidance.
 James Chung-How and David Clewer, research staff within the Centre for
Communications Research (CCR) at Bristol University, for their insights and assistance
into use of the H.263+ video codec software, and specifically to James for providing the
software utility to derive PSNR measurements.
 Tony Manuel and Richard Davies, technical managers at Lucent Technologies, for
allowing me to undertake this MSc course at Bristol University.
AUTHOR’S DECLARATION
I declare that all the work presented in this report was produced by myself, unless otherwise
stated. Recognition is made to my project supervisor, David Redmill, for his guidance. Any
use of ideas of other authors has been fully acknowledged by use of references.
Signed,
Justin Johnstone
DISCLAIMER
The views and opinions expressed in this report are entirely those of the author and not the
University of Bristol.

iii
TABLE OF CONTENTS
CHAPTER 1 ...... INTRODUCTION...................................................................................... 1
1.1. Motivation ............................................................................................................... 1
1.2. Scope of Investigations............................................................................................ 2
1.3. Organisation of this Report...................................................................................... 2
CHAPTER 2 ...... REVIEW OF THEORY AND STANDARDS .......................................... 3
2.1. Video Coding Concepts........................................................................................... 3
2.1.1. Digital Capture and Processing..................................................................................................3
2.1.2. Source Picture Format................................................................................................................4
2.1.3. Redundancy and Data Compression ..........................................................................................5
2.1.4. Coding Categories......................................................................................................................8
2.1.5. General Coding Techniques.......................................................................................................8
2.1.5.1. Block Structuring...............................................................................................................8
2.1.5.2. Intraframe and interframe coding.....................................................................................10
2.1.5.2.1. Motion Estimation ........................................................................................................10
2.1.5.2.2. Intraframe Predictive Coding........................................................................................11
2.1.5.2.3. Intra Coding..................................................................................................................12
2.1.6. Quality Assessment..................................................................................................................14
2.1.6.1. Subjective Assessment.....................................................................................................15
2.1.6.2. Objective/Quantitative Assessment..................................................................................15
2.2. Nature of Errors in Wireless Channels.................................................................. 17
2.3. Nature of Packet Errors ......................................................................................... 18
2.4. Impact of Errors on Video Coding ........................................................................ 19
2.5. Standards overview................................................................................................ 20
2.5.1. Selection of H.263+.................................................................................................................20
2.5.2. H.263+ : “Video Coding for Low Bit Rate Communications” ................................................21
2.5.2.1. Overview..........................................................................................................................21
2.5.2.2. SNR scalability ................................................................................................................21
2.5.2.3. Temporal scalability.........................................................................................................22
2.5.2.4. Spatial scalability.............................................................................................................22
2.5.2.5. Test Model Rate Control Methods...................................................................................23
2.5.3. HiperLAN/2.............................................................................................................................23
2.5.3.1. Overview..........................................................................................................................23
2.5.3.2. Physical Layer..................................................................................................................24
2.5.3.2.1. Multiple data rates ........................................................................................................25
2.5.3.2.2. Physical Layer Transmit Sequence ...............................................................................26
2.5.3.3. DLC layer ........................................................................................................................27
2.5.3.3.1. Data transport functions................................................................................................27
2.5.3.3.2. Radio Link Control functions........................................................................................28
2.5.3.4. Convergence Layer (CL)..................................................................................................28
CHAPTER 3 ...... REVIEW OF EXISTING PACKET VIDEO TECHNIQUES................... 29
3.1. Overview ............................................................................................................... 29
3.2. Techniques Considered.......................................................................................... 30
3.2.1. Layered Video and Rate Control .............................................................................................30
3.2.2. Packet sequence numbering and Corrupt/Lost Packet Treatment ............................................32
3.2.3. Link Adaptation .......................................................................................................................33
3.2.4. Intra Replenishment .................................................................................................................35
3.2.5. Adaptive Encoding with feedback ...........................................................................................35
3.2.6. Prioritised Packet Dropping.....................................................................................................36
3.2.7. Automatic Repeat Request and Forward Error Correction ......................................................36
3.2.7.1. Overview..........................................................................................................................36
3.2.7.2. Examples..........................................................................................................................37
3.2.7.2.1. H.263+ FEC..................................................................................................................37
3.2.7.2.2. HiperLAN/2 Error Control ...........................................................................................37
3.2.7.2.3. Reed-Solomon Erasure (RSE) codes ............................................................................39
3.2.7.2.4. Hybrid Delay-Constrained ARQ...................................................................................39
3.2.8. Unequal Packet Loss Protection (UPP) ...................................................................................39

iv
3.2.9. Improved Synchronisation Codewords (SCW)........................................................................40
3.2.10. Error Resilient Entropy Coding (EREC)..................................................................................41
3.2.11. Two-way Decoding..................................................................................................................41
3.2.12. Error Concealment...................................................................................................................42
3.2.12.1. Repetition of Content from Previous Frame ....................................................................42
3.2.12.2. Content Prediction and Replacement ...............................................................................42
CHAPTER 4 ...... EVALUATION OF TECHNIQUES AND PROPOSED APPROACH .... 43
4.1. General Packet Video Techniques to apply........................................................... 43
4.2. Proposal: Layered video over HiperLAN/2........................................................... 44
4.2.1. Proposal Details.......................................................................................................................45
4.2.2. Support for this approach in the HiperLAN/2 Simulation .......................................................48
CHAPTER 5 ...... TEST SYSTEM IMPLEMENTATION..................................................... 49
5.1. System Overview................................................................................................... 49
5.2. Video Sequences.................................................................................................... 49
5.3. Video Codec .......................................................................................................... 50
5.4. Scaling Function.................................................................................................... 50
5.5. HIPERLAN/2 Simulation System ......................................................................... 51
5.5.1. Existing Simulations ................................................................................................................51
5.5.1.1. Hardware Simulation .......................................................................................................51
5.5.1.2. Software Simulations .......................................................................................................51
5.5.2. Development of Bursty Error Model .......................................................................................51
5.5.3. Hiperlan/2 Simulation Software...............................................................................................53
5.5.3.1. HiperLAN/2 Error Model ................................................................................................53
5.5.3.2. HiperLAN/2 Packet Error Treatment and Reassembly ....................................................53
5.6. Measurement system ............................................................................................. 54
5.7. Proposed Tests....................................................................................................... 54
5.7.1. Test 1: Layered Video with UPP over HiperLAN/2 ................................................................54
5.7.2. Test 2: Errored Packet and Frame Treatment ..........................................................................55
5.7.3. Test 3: Performance under burst and random errors ................................................................56
CHAPTER 6 ...... TEST RESULTS AND DISCUSSION...................................................... 57
6.1. Test 1: Layered Video with UPP over HiperLAN/2.............................................. 57
6.1.1. Results......................................................................................................................................57
6.1.2. Observations and Discussion ...................................................................................................57
6.2. Test 2: Errored packet and frame Treatment ......................................................... 59
6.2.1. Results......................................................................................................................................59
6.2.2. Discussion................................................................................................................................59
6.3. Test 3: Comparison of performance under burst and random errors..................... 60
6.3.1. Results......................................................................................................................................60
6.3.2. Discussion................................................................................................................................60
6.4. Limitations in Testing............................................................................................ 61
CHAPTER 7 ...... CONCLUSIONS AND RECOMMENDATIONS..................................... 62
7.1. Conclusions ........................................................................................................... 62
7.2. Recommendations for Further Work..................................................................... 63
REFERENCES .. .................................................................................................................... 65
APPENDICES ... .................................................................................................................... 67
APPENDIX A : Overview of H263+ Optional Modes................................................... 68
APPENDIX B : Command line syntax for all simulation programs .............................. 69
APPENDIX C : Packetisation and Prioritisation Software – Source code listings........ 71
APPENDIX D : Hiperlan/2 – Analysis of Error Patterns............................................... 77
APPENDIX E : Hiperlan/2 Error Models and Packet Reassembly/Treatment software 78
APPENDIX F : Summary Reports from HiperLAN/2 Simulation modules................ 107
APPENDIX G : PSNR calculation program ................................................................ 110
APPENDIX H : Overview and Sample of Test Execution .......................................... 111

v
APPENDIX I : UPP2 and UPP3 – EP derivation, Performance comparison............... 112
APPENDIX J : Capacities of proposed UPP approach versus non-UPP approach .... 113
APPENDIX K : Recovered video under errored conditions ........................................ 114
APPENDIX L : Electronic copy of project files on CD............................................... 115

vi
LIST OF TABLES
Table 2-1 : Source picture format...........................................................................................................................5
Table 2-2 : Redundancy in digital image and video................................................................................................5
Table 2-3 : Compression Considerations................................................................................................................7
Table 2-4 : Video Layer Hierarchy.........................................................................................................................9
Table 2-5 : Distinctions between Intraframe and Interframe coding.....................................................................10
Table 2-6 : Intraframe Predictive Coding .............................................................................................................11
Table 2-7 : Wireless Propagation Issues...............................................................................................................17
Table 2-8 : Impact of Errors in Coded Video ......................................................................................................19
Table 2-9 : HiperLAN/2 Protocol Layers .............................................................................................................24
Table 2-10 : OSI Protocol Layers above HiperLAN/2 .........................................................................................24
Table 2-11 : HiperLAN/2 PHY modes.................................................................................................................25
Table 2-12 : HiperLAN/2 Channel Modes ...........................................................................................................27
Table 2-13 : Radio Link Control Functions..........................................................................................................28
Table 3-1 : General Strategies for Video Coding .................................................................................................29
Table 3-2 : Error Protection Techniques Considered ...........................................................................................30
Table 3-3 : Lost/Corrupt Packet and Frame Treatments Considered....................................................................32
Table 3-4 : Link Adpatation Algorithm – Control Issues .....................................................................................33
Table 3-5 : HiperLAN/2 Error Control Modes for user data ................................................................................38
Table 3-6 : Two-way Decoding ............................................................................................................................42
Table 4-1 : Packet Video Techniques – Extent Considered in Testing.................................................................44
Table 4-2 : Proposed Prioritisation.......................................................................................................................45
Table 4-3 : Default UPP setting for DUC of a point-to-point connection.............................................................45
Table 4-4 : Default UPP setting for DUC of a broadcast downlink connection ...................................................46
Table 4-5 : Default UPP setting for DUC of a point-to-point connection.............................................................48
Table 5-1 : Error Models for Priority Streams......................................................................................................53
Table 5-2 : Packet and frame treatment Options...................................................................................................54
Table 5-3 : Test Configurations For UPP Tests....................................................................................................55
Table 5-4 : Tests for Packet and Frame Treatment...............................................................................................55
Table 5-5 : Common Test Configurations For Packet and Frame Treatment .......................................................56
Table 5-6 : Test for Random and Burst Error Conditions.....................................................................................56
Table 5-7 : Common Test Configurations For Random and Burst Errors ............................................................56
Table 6-1 : Allocation of PHY Modes..................................................................................................................58
LIST OF ILLUSTRATIONS
Figure 2-1 : Image Digitisation...............................................................................................................................4
Figure 2-2 : Generic Video System.........................................................................................................................5
Figure 2-3 : Categories of Image and Video Coding .............................................................................................8
Figure 2-4 : Video Layer Decomposition ...............................................................................................................9
Figure 2-5 : Predictive encoding relationship between I-, P- and B- frames ........................................................10
Figure 2-6 : Motion Estimation.............................................................................................................................11
Figure 2-7 : Block Encode/Decode Actions .........................................................................................................12
Figure 2-8 : Transform Coefficient Distribution and Quantisation Matrix ...........................................................13
Figure 2-9 : SNR scalability .................................................................................................................................22
Figure 2-10 : Temporal scalability........................................................................................................................22
Figure 2-11 : Spatial scalability............................................................................................................................23
Figure 2-12 : HiperLAN/2 PHY mode link requirements.....................................................................................25
Figure 2-13 : HiperLAN/2 Physical Layer Transmit Chain..................................................................................26
Figure 2-14 : HiperLAN/2 “Packet” (long PDU) format......................................................................................28
Figure 3-1 : HiperLAN/2 Link Throughput (channel Model A)...........................................................................34
Figure 4-1 : Proposed Admission Control/QoS ....................................................................................................47
Figure 5-1 : Test System Overview ......................................................................................................................49
Figure 5-2 : Burst Error Model.............................................................................................................................52
Figure 5-3 : HiperLAN/2 Simulation....................................................................................................................53
Figure 6-1 : Comparison of EPP with proposed UPP approach over HiperLAN/2. .............................................57
Figure 6-2 : Comparison of Errored Packet and Frame Treatment.......................................................................59
Figure 6-3 : Performance under Random and Burst Error Conditions..................................................................60

1
Chapter 1 Introduction
1.1. Motivation
There is significant growth today in the area of mobile wireless data connectivity, ranging
from home and office networks to public networks, such as the emerging third generation
mobile network. In all these areas, there is expectation of an increase in support for
multimedia rich services, especially since increased data rates make this more feasible. While
there is strong incentive to develop and provide such services, the network owner/operator is
faced with many commercial and technical considerations to balance, including the following:
Consideration 1. Given a finite radio spectrum resource, there is a need to balance the
increasing bandwidth demands of any one single service against the requirement to support
as many concurrent services/subscribers as possible without causing network congestion or
overload.
Consideration 2. The need to cater for user access devices with varying complexities and
capabilities (for example, from laptop PCs to PDAs and mobile phones).
Consideration 3. The need to support end-to-end services, which may be conveyed across
numerous heterogenous networks and technology standards. For example, data originating
in a corporate Ethernet network may be sent simultaneously to both:
a) a mobile phone across a public mobile network.
b) a PC on a home wireless LAN via a broadband internet connection.
Consideration 4. The need to provide services which can adapt and compensate for the error
characteristics of a wireless mobile environment.
Digital video applications are expected to form a part of the increase in multimedia services.
Video conferencing is just one example of real-time video applications that places great
bandwidth and delay requirement demands on a network. In fact, [21] asserts that most real-
time video applications exceed the bandwidth available with existing mobile systems, and
that innovations in video source and channel coding are therefore required alongside the
advances in radio and air interface technology.
Within video compression and coding research, there has been much interest in scalable video
coding techniques. Chapter 23 of [21] defines video scalability as “the property of the
encoded bitstream that allows it to support various bandwidth constraints or display
resolutions”. This facility allows video performance to be optimised depending on both
bandwidth availability and decoder capability, albeit at some penalty to compression
efficiency. In the case of broadcast transmission over a network, a rate control mechanism
can be employed between video encoder and decoder that dynamically adapts to the
bandwidth demands on multiple links. The interest in scalable video is clearly justified by the
scenario where it is desirable to avoid compromising the quality of decoded video for one
user with high bandwidth and high specification user equipment (e.g. a PC via a broadband
internet connection), while still providing a reduced quality video to a second user with low
bandwidth and low specification equipment (e.g. a mobile phone over a public mobile
network). Another illustration of this problem is given in [19], which highlights the similar
predicament currently facing the web design community. With the more recent introduction
of web-connected platforms, such as PDAs and WAP phones, web designers are now
challenged to produce web content which can dynamically cater for varying levels or classes
of access device capability (e.g. screen size and processing power). One potential strategy

2
identified in [19] echoes the video scalability principle, in that ideally, the means to derive
“reduced complexity” content should be embedded within each web page. This would allow
web designers to avoid the undesired task of publishing and maintaining multiple sites - each
dedicated to the capabilities of each class of access device. With the potential for significant
growth in video applications via broadband internet and third generation mobile access, it is
therefore inevitable that for the same reason, scalable video will become increasingly
important too.
1.2. Scope of Investigations
The brief of this project is to focus on optimising aspects of scalable video coding over a
wireless LAN, and specifically the HiperLAN/2 standard [2]. Within the HiperLAN/2
standard, a key feature of the physical layer is that it supports a number of different physical
modes, which have differing data rates and link quality requirements. Specifically, the modes
with higher nominal data rates require higher carrier-to-interference (C/I) levels to be able to
achieve near their nominal data rates. Performance studies ([7],[16],[20]) have shown that as
the C/I decreases, higher overall throughput can be maintained if the physical mode is
reduced at specific threshold points. The responsibility to monitor and dynamically alter
physical modes is designated to a function referred to as “link adaptation”. However, as this is
function is not defined as part of the HiperLAN/2 standard, there is interest to propose
algorithms in this area. This project was tasked to recommend an optimal scheme to convey
layered video data over HiperLAN/2, including assessment of link adaptation algorithms.
However, when making proposals in any one specific area of mobile multimedia applications,
it is vital to consider these in the wider context of the entire system in which they will
operate. Therefore this report also considers many of the more general techniques in the area
of “wireless packet video”.
1.3. Organisation of this Report
The remainder of this report is organised as follows: Chapter 2 commences with an
introduction of video coding theory. It then provides an overview of nature of errors in a
wireless environment, and what effect these have on the performance of video coding.
Chapter 2 concludes with an overview of the standards used in the project, namely H.263+
and HiperLAN/2. Chapter 3 presents a review of some existing wireless packet video
techniques, which are considered for investigation in this project. Chapter 4 proposes the
approach intended to convey layered video bitstream over HiperLAN/2. The chapter also
considers which of the more general techniques to apply during testing. Chapter 5 describes
the implementation of components in the simulation system used for investigations, and then
introduces the specific tests to be conducted. Chapter 6 discusses test results, and chapter 7
presents conclusions, including a list of areas identified for further investigation.

3
Chapter 2 Review of Theory and Standards
This chapter introduces some useful concepts regarding video coding, wireless channel errors
and their effect on coded video. The chapter concludes with an overview of the standards
used in the project; H.263+ and HiperLAN/2.
2.1. Video Coding Concepts
The following basic concepts are discussed in this section:
 Digital image capture and processing considerations
 Picture Formats
 Redundancy and Compression
 Video Coding Techniques
 Quality assessment
The video coding standards considered for use on this project include H.263+, MPEG-2 and
MPEG-4, since all these standards incorporate scalability. While the basic concepts discussed
in this section are common to all these standards, many of the illustrations are tailored to
highlight specific details of H.263+, since this is the standard selected for use in this project.
Further details of this standard, including it scalability modes, are discussed later in section
2.5.1.
2.1.1. Digital Capture and Processing
Digital images provide a two dimensional representation of the visual perception of the three-
dimensional world. As indicated in [22], the process to produce digital images involves the
following two stages, shown also in Figure 2-1 below:
 Sampling: Here an image is partitioned into a grid of discrete regions called
picture elements (abbreviated “pixels” or “pels”). An image capture
device obtains measurements for each pixel. Typically this is performed
by a scan, which commences in one corner of the grid, and proceeds
sequentially row by row, and pixel by pixel within each row, until
measurements for the last pixel, in the diagonally opposite corner to the
start point are recorded. Commonly, image capture devices record the
colour components for each pixel in RGB colour format.
 Quantisation: Measured values for each pixel are mapped to the nearest integer within
a given range (typically this range is 0 to 255). In this way, digitisation
results in an image being represented by an array of integers.
Note:. The RGB representation is typically converted into YUV format
(where Y represents luminance and U,V represent chrominance
values). The significant benefit of this conversion is that it
subsequently allows more efficient coding of the data.

4
Figure 2-1 : Image Digitisation
Digital video is a sequence of still images (or frames) displayed sequentially at a given rate,
measured in frames per second (fps). When designing a video system, trade-off decisions
need to be made between the perceived visual quality of the video and the complexity of the
system, which may typically be characterised by a combination of the following:
 Storage Space/cost
 Bandwidth Requirements (i.e. data rate required for transmission)
 Processing power/cost.
These trade-off decision can be partly made by defining the following parameters of the
system. Note that increasing these parameters enhances visual quality at the expense of
increased complexity.
 Resolution: This includes both:
 The number of chrominance/luminance samples in horizontal and
vertical axes.
 The number of quantised steps in the range of chrominance and
luminance values (for example, whether 8, 16 or more bits are
stored for each value).
 Frame rate: The number of frames displayed per second.
Sensitivity limitations of the Human Visual System (HVS) effectively establish practical
upper thresholds for the above parameters, beyond which there are diminishing returns in
perceived quality. An example of video coding adapting to the HVS limits is the technique
known as “chrominance sub-sampling”. Here, the number of chrominance samples is reduced
by a given factor (typically 2) relative to the number of luminance samples in one or both of
the horizontal and vertical axes. The justification for this is that the HVS is spatially less
sensitive to differences in colour than brightness.
2.1.2. Source Picture Format
Common settings for resolution parameters have been defined within video standards. One
example is the “Common Intermediate Format” (CIF) picture format. The parameters for CIF
are shown in Table 2-1 below, along with some common variants.

5
Table 2-1 : Source picture format
Source Picture Format Resolution
(Number of samples in
horizontal x vertical)
Luminance Chrominance
16CIF 1408x1152 704x576
4CIF 704x576 352x288
CIF 352 x 288 176 x 144
QCIF (Quarter-CIF) 176 x 144 88 x 72
sub-QCIF 128x96 64x48
2.1.3. Redundancy and Data Compression
Even with fixed setting for resolution and frame rate parameters, it is both desirable and
possible to further reduce the complexity of the video system. Significant reductions in
storage or bandwidth requirements may be achieved by encoding and decoding of the video
data to achieve data compression. A generic video coding scheme depicting this compression
is shown in Figure 2-2 below.
reference
input video
recovered
output
video
transmission
channel
or
data storage
Video
Encoder
compressed
data
compressed
data Video
Decoder
Figure 2-2 : Generic Video System
Data compression is possible, since video data contains redundant information. Table 2-2 lists
the types of redundancy and introduces two compression techniques which take advantage of
this redundancy. These techniques are discussed later in section 2.1.5.
Table 2-2 : Redundancy in digital image and video
Domain
where
Redundancy
Exists
Description Applicable
Compression
Technique
Spatial: Within a typical still image, spatial redundancy exists in a
uniform region of the image, where luminance and
chrominance values are highly correlated between
neighbouring samples.
Intraframe
Predictive
coding
Temporal: Video contains temporal redundancy in that data in one
frame is often highly correlated with data in the same
region of the subsequent frame (e.g. where the same
object appears in the subsequent frames, albeit with some
motion effects). This correlation diminishes as the frame
rate is reduced and/or the motion content within the video
sequence increases.
Interframe
Predictive
coding.

6
While the encoding and decoding procedures do increase processing demands on the system,
this penalty is often accepted when compared to the potential savings in storage and
bandwidth costs. Apart from increased processing power, there are other options to consider
during compression. These generally trade-off compression efficiency against perceived
visual quality, as shown in Table 2-3 below.

7
Table 2-3 : Compression Considerations
Consideration Factor Effects on:
Compression Efficiency Perceived Quality
a) Resolution, Frame Rate  Skipping frames/ reducing
the frames rate will
increase efficiency.
 Digitising the same
picture with an increased
number of pixels allows
that more spatial
correlation and
redundancy can be
exploited during
compression.
 Increased resolution (e.g.
going from 8 to 16 bit
representation) will
decrease efficiency.
Increases in resolution and
frame rate typically
improve quality at the
expense of increased
processing
complexity/cost.
b) Latency:
end-to-end delays
introduced into the
system due to finite
processing delays.
Improved compression may
improve or degrade latency,
dependant on other system
features. At one extreme, it
may increase latency due to
increased complexity and
therefore increased processing
delays. At the other extreme,
in a network, it may result in a
reduced bandwidth demand.
This may serve to reduce
network load and actually
reduce buffering delays and
latency in the network.
If frames are not played
back at a regular frame
rate, the resultant pausing
or skipping will drastically
reduce perceived quality.
c) Lossless compression:
where the original can
be reconstructed exactly
after decompression.
This category of compression
typically achieves lower
compression efficiency than
lossless compression.
By definition, the
recovered image is not
degraded. This may be
more applicable to medical
imaging or legal
applications where storage
costs or latency are of less
concern.
d) Lossy compression:
where the original
cannot be reconstructed
exactly.
This typically achieves higher
compression efficiency than
lossless.
By definition, the quality
will be degraded, however
techniques here may
capitalise on HVS
limitations, so that the
losses are less perceivable.
This is more applicable to
consumer video
applications.

8
2.1.4. Coding Categories
There is a wide selection of strategies for video coding. The general categories into which
these strategies are allocated is shown in Figure 2-3 below. As indicated by the shading in the
diagram, this project makes use of DCT transform coding, which is categorised as a lossy,
waveform based method.
Categories of Image and Video Coding
MODEL based WAVEFORM based
Statistical Universal Spatial Frequency
LossyLoss-less
example:
- wire frame modelling
examples:
- sub-band
examples:
- fractal
- predicitive(DPCM)
example:
- ZIP
examples:
- Huffman
- Arithmetic - DCT transform
Note: shaded selections identify the categories and method considered in this project.
Figure 2-3 : Categories of Image and Video Coding
2.1.5. General Coding Techniques
To achieve compression, a video encoder typically performs a combination of the following
techniques, which are discussed briefly in the sub-sections below this list:
 Block Structure and Video Layer Decomposition
 Intra and Inter coding: Intraframe prediction, Motion Estimation
 Block Coding: Transform, Quantisation and Entropy Coding.
2.1.5.1. Block Structuring
The raw array of data samples produced by digitisation is typically restructured into a format
that the video codec can take advantage of during processing. This new structure typically
consists of a four layer hierarchy of sub-elements. A graphical view of this decomposition of
layers, with specific values for QCIF and H.263+ is given in Figure 2-4 below. When coding
schemes apply this style of layering, with the lowest layer consisting of blocks, the coding
scheme is said to be “block-based” or “block-structured”. Alternatives to block-based coding
are object- and model-based coding. While these alternatives promise the potential for even
lower bit rates, they are less mature than current block-based techniques. These will not be
covered in this report, however more information can be found in chapter 19 of [21].

9
Figure 2-4 : Video Layer Decomposition
A brief description of the contents of these four layers is given in Table 2-4 below. Some of
the terms introduced in this table are be covered later.
Table 2-4 : Video Layer Hierarchy
Layer Description/Contents
a) Picture: This layer commences each frame in the sequence, and includes:
 Picture Start Code.
 Frame number.
 Picture type, which is very important as it indicates how to
interpret subsequent information in the sub-layers.
b) Group of blocks
(GOB):
 GOB start code.
 Group number.
 Quantisation information.
c) MacroBlock (MB):  MB address.
 Type (indicates intra/inter coding, use of motion
compensation).
 Quantisation information.
 Motion vector information.
 Coded Block Pattern.
d) Block:  Transform Coefficients.
 Quantiser Reconstruction Information.

10
2.1.5.2. Intraframe and interframe coding
The features and main distinctions between these two types of coding is summarised briefly
in Table 2-5 below.
Table 2-5 : Distinctions between Intraframe and Interframe coding
Coding Style Features and Implications
INTER coding  Frames coded this way rely on information in previous frames, and
code the relative changes from this reference frame.
 This coding improves compression efficiency (compared to intra
coding).
 The decoder is only able to reconstruct an inter coded frame when it has
received both: a) the reference frame, and b) the inter frame.
 The common type of inter coding, relevant to project, is ‘Motion
Estimation and Compensation’, which is discussed below.
 The reference frame(s) can be forward or backwards in time (or a
combination of both) relative to the predictive encoded frame. When
the reference frame is a previous frame, the encoded frame is called a
“P-frame”. When reference frames are previous and subsequent frames,
the encoded frame is termed a “B-frame” (Bi-directional). This
relationship is made clear in Figure 2-5 below.
 Since decoding depends on data in reference frames, any errors that
occur in the reference will persist and propagate into subsequent frames
– this is known as “temporal error propagation”.
INTRA coding  Does not rely on previous (or subsequent) frames in the video
sequence.
 Compression efficiency is reduced (compared to inter coding).
 Recovered video quality is improved. Specifically, if any temporal error
propagation existed, it will be removed at the point of receiving the
intra coded frame.
 Frames coded this way are termed “I-frames”.
I-frame
I-frame
P-frame
B-frame
P-frame
B-frame
Forward Prediction from Reference
Bi-Directional Prediction from References
time
Figure 2-5 : Predictive encoding relationship between I-, P- and B- frames
2.1.5.2.1. Motion Estimation
Motion estimation and compensation techniques provide data compression by exploiting
temporal redundancy in a sequence of frames. “Block Matching Motion Estimation” is one
such technique, which results in a high levels of data compression.

11
The essence of this technique is that for each macroblock in the current frame, an attempt is
made to locate a matching macroblock within a limited search window in the previous frame.
If a matching block is found, the relative spatial displacement of the macroblock between
frames is determined, and this is coded and sent as a motion vector. This is shown in Figure
2-6 below. Transmitting the motion vector instead of the entire block represents a vastly
reduced amount of data. When no matching block is found, then the block will need to be
intra coded. Limitations of this technique include the following:
 It does not cater for three dimensional aspects such as zooming, object rotation and object
hiding.
 It suffers from temporal error propagation.
matched
MB n
(x+ Dx, y+Dy)
Macroblock n
position in
previous frame
(x,y)
Macroblock n
position in
current frame
MB n
(Dx, Dy)
motion vector
for MB n
search window -
vertical search range
search window -
horizontal search range
Figure 2-6 : Motion Estimation
2.1.5.2.2. Intraframe Predictive Coding
This technique provides compression by taking advantage of the spatial redundancy within a
frame. This technique operates by performing the sequence of actions at the encoder and
decoder as listed in Table 2-6 below.
Table 2-6 : Intraframe Predictive Coding
Actions
performed
by:
Actions for each sample:
1. Encoder: a) An estimate for the value of the current sample is produced by a
prediction algorithm that uses values from ‘neighbouring” samples.
b) The difference (or error term) between the actual and the predicted
values is calculated.
c) The error term is sent to the decoder (see notes 1 and 2 below).
2. Decoder: a) The same prediction algorithm is performed to produce an estimate for
the current sample.
b) This estimate is adjusted by the error term received from the encoder to
produce an improved approximation of the original sample.

12
Notes:
1) For highly correlated samples in a uniform region of the frame, the prediction will be very
precise, resulting in a small or zero error term, which can be coded efficiently.
2) When edges occur between non-uniform regions within an image, neighbouring samples
will be uncorrelated and the prediction will produce a large error term, which diminishes
coding efficiency. However, chapter 2 of [23] highlights the following two factors which
determine that this technique does provide overall compression benefits:
a) Within a wide variety of naturally occurring image data, the vast majority of error
terms are very close to zero – hence on average, it is possible to code error terms
efficiently.
b) While the HVS is very sensitive to the location of edges, it is not as sensitive the
actual magnitude change. This allows that the large magnitude error terms can be
quantised coarsely, and hence coded more efficiently.
3) Since this technique only operates on elements within the same frame, the “intraframe”
aspect of it’s name is consistent with the definitions in Table 2-5. However, since coded
data depends on previous samples, it is subject to error propagation in a similar way to
interframe coding. This error propagation may manifest itself as horizontal and/or vertical
streaks to the end of a row or column.
2.1.5.2.3. Intra Coding
When a block is intra coded (for example, when a motion vector cannot be used), the encoder
translates the data in the block into a more compact representation before transmitting it to
the decoder. The objective of such translation (from chapter 3 of [23]) is: “to alter the
distribution of the values representing the luminance levels so that many of them can either be
deleted entirely, or at worst be quantised with very few bits”.
The decoder must perform the inverse of the translation actions to restore the block back to
the original raw format. If the coding method is categorised as being loss-less, then the
decoder will be able to reconstruct the original data exactly. However, this project makes use
of lossy a coding method, where a proportion of the data is permanently lost during coding.
This restricts the decoder to reconstruct an approximation of the original data.
The generic sequence of actions involved in block encoding and decoding are shown in
Figure 2-7 below. Each of these actions is discussed in the paragraphs below, which also
introduce some specific examples relevant to this project.
Transform
block
(e.g. 8x8)
Entropy
Code
Quantise
Inverse
Transform
Requantise
transmission
channel
or storage
Entropy
Decode
ENCODER
block
(e.g. 8x8)
DECODER
Figure 2-7 : Block Encode/Decode Actions

13
Transformation
This involves translation of the sampled data values in a block into another domain. In the
original format, samples may be viewed as being in the spatial domain. However, chapter 3 of
[23] indicates these may equally be considered to be in the temporal domain, since the
scanning process during image capture provides a deterministic mapping between the two
domains. The coding category highlighted in Figure 2-3 for use in this project is
“Waveform/Lossy/Frequency”. The frequency aspect refers to the target domain for the
transform. The transform can be therefore be viewed as a time to frequency domain
transform, allowing selection of a suitable Fourier Transform based method. Possibilities for
this include:
 Walsh-Hadamard Transform (WDT)
 Discrete Cosine Transform (DCT)
 Karhunen-Loeve Transform (KLT)
The DCT is a popular choice for many standards, including JPEG and H.263. The output of
the transform on an 8x8 (for example) block results in the same number (64) of transform
coefficients, each representing the magnitude of frequency component. The coefficients are
arranged in the block with the single DC coefficient in the upper left corner, then zig-zag
along diagonals with the highest frequency coefficient in the lower right corner of the block
(see Figure 2-8). Since the transform produces the same number of coefficients as there were
to start with; it does not directly achieve any compression. However, compression is achieved
in the subsequent step, namely quantisation. Further details of the DCT transform are not
presented here, as these are not directly relevant to this project, however [23] may be referred
to for more details.
Quantisation
This operation is performed on the transform coefficients. These coefficients usually have
widely varying range of magnitudes, with a typical distribution as shown in the left side of
Figure 2-8 below. This distribution makes it possible to ignore some of the coefficients with
sufficiently small magnitudes. This is achieved by applying a variable quantisation step size
to positions in the block, as shown in the right side of Figure 2-8 below. This approach
provides highest resolution to the DC and high magnitude coefficients in the top left of the
block, while reducing resolution toward the lower right corner of the block. The output of the
quantisation typically results in a block with some non-zero values towards the upper left of
the block, but many zeros towards the lower right corner of the block. Compression is
achieved when these small or zero magnitude values are coded more efficiently as seen in the
subsequent step, namely Entropy Coding.
DC
large magnitude
low frequency
coefficients
small magnitude
high frequency
coefficients
8x8 block of transform coefficients
smallest
quantistion values
- better resolution
largest
quantisation values,
lowest resolution
corresponding 8x8 matrix of quantisation
coefficients
increasing
quantisation
values
increasing
quantisation
values
coefficients
loaded
diagnoanally
Figure 2-8 : Transform Coefficient Distribution and Quantisation Matrix

14
Entropy Coding
This type of source coding provides data compression without loss of information. In general,
source coding schemes map each source symbol to a unique (typically binary) codeword prior
to transmission. Compression is achieved by coding as efficiently as possible (i.e. using the
fewest number of bits). Entropy coding achieves efficient coding by using variable length
codewords. Shorter codewords are assigned to those symbols that occur more often, and
longer codewords are assigned to symbols that occur less often. This means that on average,
shorter codewords will be used more often, resulting in fewer bits being required to represent
an average sequence of source symbols.
An optional, but desirable property for variable length codes is the “prefix-free” property.
This property is satisfied when no codeword is the prefix of any other codeword. This has the
benefit that codewords are “instantaneously decodable”, which implies that the end of a
codeword can be determined as the codeword is read (and without the need to receive the start
of the next codeword). Unfortunately, there are also some drawbacks in using variable length
codes. These include:
1) The instantaneous rate at which codewords are received will vary due to the variable
codeword length. A decoder therefore requires buffering to smooth the rate at which
codewords are received.
2) Upon receipt of a corrupt codeword, the decoder may lose track of the boundaries
between codewords. This can result in the following two undesirable outcomes, which are
relevant as their occurrence is evident later in this report:
a) The decoder may only recognise the existence of an invalid codeword sometime after
it has read beyond the start of subsequent codewords. In this way, a number of valid
codewords may be incorrectly ignored until the decoder is able to resynchronise itself
to the boundaries of valid codewords.
b) When attempting to resynchronise, a “false synchronisation” may occur. This will
result in further loss of valid codewords until the decoder detects this condition and is
able to establish valid resynchronisation.
An example of a prefix-free variable length code is the Huffman Code. This has the
advantage that it “can achieve the shortest average code length” (chapter 11 of [24]). Further
details of Huffman coding are not covered here, as they are not of specific interest to this
project.
Within a coded block of data, it common to observe a sequence of repeated data values. For
example, the quantised transform coefficients near the lower right corner of the block are
often reduced to zero. In this case, it is possible to modify and further improve the
compression efficiency of Huffman (or other source) coding, by applying “Run-Length
coding”. Instead of encoding each symbol in a lengthy run of repeated values, this technique
describes the run with an efficient substitution code. This substitution code typically consists
of a control code and a repetition count to allow the decoder to recreate the sequence exactly.
2.1.6. Quality Assessment
When developing video coding systems, it is necessary to provide an assessment of the
quality of the recovered decoded video. Such assessments provide performance indicators
which are necessary to optimise system performance during development or even to confirm
the viability of the system prior to deployment. It is usually desirable to perform subjective
and quantitative assessments of video performance. It is quite common to perform
quantitative assessments during earlier development and optimisation phases, and then to

15
conduct subjective testing prior to deployment. Features of these two types of assessment are
discussed below.
2.1.6.1. Subjective Assessment
Testing is performed by conducting formal human observation trials with a statistically
significant population of the target audience for the video application. During the trials,
observers independently view video sequences and then rate their opinion of the quality
according to a standardised five point scoring system. Scores from all observers are then
averaged to provide a “Mean Opinion Score” (MOS) for each video sequence.
A significant advantage of subjective assessment trials is that they allows observers to
consider the entire range of psycho-visual impressions when rating a test sequence. Trial
results therefore reflect a full consideration of all aspects of video quality. MOS scores are
therefore a well accepted means to compare the relative performance of different systems.
The following implications of subjective assessments are noted:
 Formal trials are relatively costly and time consuming. This can result in trials being
skipped or deferred to later stages of system development.
 Performance of the same system can vary widely depending on the specific video content
used in trials. In the same way, observers’ perceptions are equally dependent on the video
content. Therefore, comparison of MOS scores between different systems is only credible
if identical video sequences are used.
 Observers’ perceptions may vary dependant on the context of the environment in which
they are viewed. An illustration of this is given in mobile telephony. Here, lower MOS
scores for the audio quality of mobile telephony (when compared to higher MOS scores
for the quality of PSTN) are tolerated when taking into account the benefits of mobility.
This tolerant attitude is likely to persist toward video services offered over mobile
networks. Therefore, observation trials should account for this attitude when conducting
trials and reporting MOS scores.
 Alongside the anticipated rapid growth in mobile video services, it is likely that audience
expectations will rapidly become more sophisticated. It is therefore suggested that
comparing results between two trials is less meaningful if the trial dates are separated by a
significant time period (say 2 years, for example).
2.1.6.2. Objective/Quantitative Assessment
In strong contrast to subjective testing, where consideration of all aspects of the quality are
incorporated within a MOS score; quantitative assessments are able to focus on specific
aspects of video quality. It is therefore often more useful if these assessments are quoted in
the context of some other relevant parameters (such as frame rate or channel error rate). The
defacto measurement used in video research is Peak-Signal-to-Noise Ratio (PSNR), described
below.
Peak-Signal-to-Noise Ratio (PSNR)
This assessment provides a measure in decibels of the difference between original and
reconstructed pixels over a frame. It is common to quote PSNR for luminance values only,
even though PSNR can be derived for chrominance values as well. This project adopts the
convention of using luminance PSNR only. The luminance PSNR for a single frame is
calculated according to equation (2) below.

16









MSE
PSNR frame
2
MAX_LEVEL
10log10
(2)
Where,
MAX_LEVEL= The maximum value luminance
(when using 8 bit representation for luminance, this value
is 255)
MSE = “Mean Squared Error” of decoded pixel luminance
amplitude compared with original reference pixel, given
by equation (3)
 2
ValuePixelOrignalValuePixelRecovered
1
 
pixelsNpixelsN
MSE
(3)
Where,
Npixels = Total number of pixels in the frame.
PSNR may be also be calculated for an entire video sequence. This is simply an average of
PSNR values for each frame, as shown in equation (4) below.

framesAll
frame
frames
sequence PSNR
N
PSNR
1 (4)
Where,
N frames = Total number of decoded frames in sequence.
For example, if only frame numbers 1, 6 and 10 are
decoded,
then N frames = 3.
The following implications of using PSNR as a quality assessment are noted:
1. PSNR for a sequence can only be based on the subset of original frames that are decoded.
When compression results in frames being dropped, this will not be reflected in the PSNR
result. This is a case where quoting the frame rate alongside PSNR results is appropriate.
2. The PSNR calculation requires a reference of the original input video is required at the
receiver. This may only be possible in a controlled test environment.
3. PSNR is measured between the original (uncompressed) frames and the final decoded
output frames. This therefore reflects a measure of effects across the entire system,
including the transmission channel, which may introduce errors. Hence quoting the error
rate of the channel alongside the PSNR result may be appropriate in this case
4. Since both:
a) PSNRframe is averaged over the pixels in a frame
b) PSNRsequence is averaged over the frames in a sequence,
these averages are unable to highlight the occurrence of localised quality degradations
which may be intolerable to a viewer. Based on this, it is highly recommended that when
quoting PSNR measurement, observations from informal (as a minimum) viewing trials
are noted as well.

17
2.2. Nature of Errors in Wireless Channels
Errors in wireless channels result from propagation issues described briefly in Table 2-7
below.
Table 2-7 : Wireless Propagation Issues
Propagation
Issue
Description, Implications
1) Attenuation For a line of sight (LOS) radio transmission, the signal power at the
receiver will be attenuated due to “free space attenuation loss”, which is
inversely proportional to square of the range between transmitter and
receiver. For non-LOS transmissions, the signal power will also suffer
“penetration losses” when entering or passing through objects.
a) Multipath
&
Shadowing
In a mobile wireless environment, transmitted signals will undergo effects
such as reflection, absorption, diffraction and scattering off large objects in
the environment (such as buildings or hills). Apart from further attenuating
the signal, these effects result in multiple versions of the signal being
observed at the receiver. This is termed multipath propagation.
When the LOS signal component is lost, the effect on received signal
power is determined by combination of these multipath signals. Severe
power fluctuations at the mobile receiver result as it moves in relation to
objects in the environment. The rate of these power fluctuations is limited
by mobile speeds in relation to these objects, and is typically less than two
per second. For this reason, this power variation is termed “slow fading”.
Another type of signal fluctuation exists and this is termed “fast fading”.
Due to the differing path lengths of the multipath signals; these signals
possess different amplitudes and phases. These will result in constructive
or destructive interference when combined instantaneously at the receiver.
Minor changes in the environment or very small movements of the mobile
receiver can dynamically modify these multipath properties, and this
results in rapid fluctuations (above 50Hz, for example) of received signal
power – hence the name “fast fading”.
As slow and fast fading are occur at the same time, their effects on
received power level are superimposed. During fades, received power
levels can drops by tens of dB, which can drastically impair system
performance.
b) Delay
Spread
Another multipath effect is “delay dispersion”. Due to the varying paths
lengths of the versions of the signal, these will arrive at slightly differing
times at the receiver. If the delay spread is larger than a symbol period of
the system, energy from one symbol will be received during the next
symbol period, and this will interfere with the interpretation of that
symbol’s true value. This undesirable phenomenon is termed Inter-Symbol
Interference (ISI).
c) Doppler
effects
When a mobile receiver moves within the reception area, the phase at the
receiver will constantly change due to changing path length. This phase
change represents a frequency, known as the Doppler Frequency. At the
receiver, unless the offset of the Doppler frequency is tracked, this reduces
the energy received over a symbol period, which degrades system
performance.

18
The combination of these effects, but specifically multipath, results in received signal strength
undergoing continual fades of varying depth and duration. When the received signal falls
below a given threshold during a fade, then unrecoverable errors will occur at the receiver
until the received signal level exceeds the threshold again. Errors therefore occur in bursts
over the duration of such fades. This bursty nature of the errors in a wireless is the key point
raised in this section. Therefore, when simulating errors in a wireless transmission channel, it
is realistic to represent this by a bursty error model (as opposed to a random error model).
2.3. Nature of Packet Errors
Degradation of coded video in a packet network can be attributed to the following factors:
a) Packet Delay: Whether overall packet delay is a problem depends on the application.
One-way, non-interactive services such as video streaming or
broadcast can typically withstand greater overall delay. However, two-
way, interactive services such as video conferencing are adversely
effected by overall delay. Excessive delay can result in effective
packet loss, where, for example, if a packet arrives beyond a cut-off
time for the decoder to make use of the packet, it is discarded.
b) Packet Jitter: This relates to the amount of variation of delay. Jitter may be caused
by varying queuing delays at each node across a network. Different
prioritisation strategies for packets queues can have significant effects
on delay, especially as network congestion increases.
The effects of jitter depend on the applications, as before. One-way,
non-interactive applications can cope as long as sufficient buffering
exists to smooth the variable arrival rate of packets, so that the
application can consume packets at a regular rate. Problems still exist
for two-way, interactive services when excessive delays result in
effective packet loss.
c) Packet Loss: Corrupt packets will be abandoned as soon as they are detected in the
network. Some queuing strategies can also discard valid packets if
they are noted as being “older” than a specified temporal threshold.
Since these packets are too late to be of use, there is no benefit in
forwarding them. This technique can actually alleviate congestion in
networks and reduce overall delays. The fact that an entire packet is
lost (even if only a single bit in it was corrupt), further contributes to
the bursty nature of errors already mentioned for wireless channels.

19
2.4. Impact of Errors on Video Coding
Chapter 27 of [21] provides a good illustration of the potentially drastic impact of errors in a
video bitstream. Table 2-8 provides a summary of these impacts, showing the type of error
and it’s severity. When error propagation results, the table categorises the type of propagation
and lists some typical means to reduce the extent of the propagation. The examples in this
table are ordered from lowest to highest severity impact.
Table 2-8 : Impact of Errors in Coded Video
Example 1:
Location of error: Within a macroblock of an I-frame, which is not used as a
reference in motion estimation.
Extent of impact: The error is limited to a (macro)block within one frame only –
there is no propagation to subsequent frames. This therefore
results in a single block error.
Severity: Low
Error Propagation
Category:
Not applicable – no propagation
Example 2:
Location of error: In motion vectors of a P-frame, that does not cause loss of
synchronisation.
Extent of impact: Temporal error propagation results, with multiple block errors
in subsequent frames until the next intra coded frame is
received.
Severity: Medium
Error Propagation
Category:
Predictive Propagation
Propagation Reduction
strategy:
The accepted recovery method is to send an intra coded frame
to the decoder.
Example 3:
Location of error: In a variable length codeword of a block which is used for
motion estimation, and causes loss of synchronisation.
Extent of impact: When the decoder loses synchronisation, it will ignore valid
data until it is able to recognise the next synchronisation
pattern. If only one synch pattern is used per frame, the rest of
the frame may be lost. If the error occurred at the start of an I-
frame, not only is this frame lost, but subsequent P-frames
will also be severely errored, due to temporal propagation.
This potentially results in multiple frame errors.
Severity: High
Error Propagation
Category:
Loss of Synchronisation
strategy:
One way to reduce the extent of this propagation, is to insert
more synchronisation codewords into the bitstream so that
less data is lost before resynchronisation is gained. This has
the drawback of increasing redundancy overheads.

20
Example 4:
Location of error: In the header of a frame.
Extent of impact: Information in the header of a frame contains important
information about how data is interpreted in the remainder of
the frame. If this information is corrupt, data in the rest of the
frame is meaningless – hence the entire frame can be lost.
Severity: High
Error Propagation
Category:
Catastrophic
strategy:
Protect the header by a powerful channel code to ensure it is
received intact. Fortunately, header information represents a
small percentage of the bitstream, so the overhead associated
with protecting it is less onerous than might be expected.
Based on the potentially dramatic impact of these errors, it is evident that additional
techniques are required to protect against these errors and their effects in order to enhance the
robustness of coded video. Examples of such techniques are discussed in chapter 3.
2.5. Standards overview
2.5.1. Selection of H.263+
H.263+ was selected as the video coding standard for use on this project. Justifications for
choosing H.263+ in preference to MPEG include the following:
 H.263+ simulation software is more readily available and mature than MPEG software.
Specifically, an H.263+ software codec [3] is available within the department, and there
are research staff familiar with this software.
 The H263+ specification is much more concise than the MPEG specifications, and given
the timeframe of this project, it was considered more feasible to work with this smaller
standard.
 H.263+ is focussed at lower bit rates, which is of more interest in this project. This focus
is based on the anticipation of significant growth in video applications being provided
over the internet and public mobile networks. To make maximum use of their bandwidth
investments, mobile network operators will be challenged to provide services at lower bit
rates.
Based on this choice, some unique aspects of MPEG-2 and MPEG-4, such as those listed
below, are not covered in this project:
 Fine-Grained-Scalability (FGS).
 Higher bit rates (MPEG-2 operates from 2 to 30Mbps).
 Content based scalability (MPEG-4).
 Coding of natural and synthetic scenes (MPEG-4)

21
2.5.2. H.263+ : “Video Coding for Low Bit Rate Communications”
2.5.2.1. Overview
H.263+ refers to the version 2 issue of the original H.263 standard. This specification defines
the coded representation and syntax for compressed video bitstream. The coding method is
block-based and uses motion compensation, inter picture prediction, DCT transform coding,
quantisation and entropy coding, as discussed earlier in Chapter 2. The standard specifies
operation with all five picture formats listed in Table 2-1, namely sub-QCIF, QCIF, CIF,
4CIF and 16CIF. The specification defines the syntax in the four layer hierarchy (Picture,
GOB, MacroBlock and Block) as discussed earlier.
The specification defines sixteen negotiable coding modes in a number of annexes. A brief
description of all these modes is given in Appendix A. However, the most significant mode
for this project is given in annex O, which defines the three scalability modes supported by
the standard. With scalability enabled, the coded bitstream is arranged into a number of
layers, starting with a mandatory base layer and one or more enhancement layers, as follows:
Base Layer A single base layer is coded which contains the most important
information in the video bitstream. It is essential for this layer to be
sent to the decoder as it contains what is considered as the ‘bare
minimum’ amount of information required for the decoder to
produce acceptable quality recovered video. This layer consists of I-
frames and P-frames.
Enhancement
Layer(s)
One or more enhancement layers are coded above the base layer.
These contain additional information, which, if received by the
decoder, serve to improve the perceived quality of the recovered
video produced by the decoder. Each layer can only be used by the
decoder if all layers below it have been received. The lower layer
thus forms a reference layer for the layer above it. The modes of
scalability relate directly to the way in which they improve the
perceived quality of the recovered video, namely: resolution, spatial
quality or temporal quality. These modes are described in the
following sections. Note that it is possible to combine the scalability
modes into a hybrid scheme, although this is not considered in this
report.
2.5.2.2. SNR scalability
The perceived quality improvement derived from the enhancement layers in this mode is
pixel resolution (luminance and chrominance).
This mode provides “coding error” information in the enhancement layer(s). The decoder uses
this information in combination with that received in the base layer. After decoding
information in the base layer to produce estimated luminance and chrominance value for each
pixel, the coding error is then used to adjust these values to create a more accurate
approximation of the original values. Frames in the enhancement layer are called
Enhancement I-frames (EI) and Enhancement P-frames (EP). The “prediction direction”
relationship between I and P frames is carried over to the EI and EP frames in the
enhancement layer. This relationship is shown in Figure 2-9 below.

22
EI EP EP EP
I P P P
SNR scaled sequence
Enhancement
Layer 1
BASE
layer
key:
direction of prediction
Figure 2-9 : SNR scalability
2.5.2.3. Temporal scalability
The perceived quality improvement derived from the enhancement layers in this mode is
frame rate.
This mode provides the addition of B frames, which are bi-directionally predicted and
inserted between I and P frames. These additional frames increase the decoded frame rate, as
shown in Figure 2-10 below.
I P P P
B B B
Temporal scalability
key:
enhancement
layer:
base
layer:
Figure 2-10 : Temporal scalability
2.5.2.4. Spatial scalability
The perceived quality improvement factor derived from the enhancement layers in this mode
is picture size.
In this mode, the base layer codes the frame in a reduced size picture format (or equivalently,
a reduced resolution at the same picture format size). The enhancement layer contains
information to increase the picture to its original size (or resolution). Similar to the SNR
scalability, the enhancement layer contains EI and EP frames with the same prediction
relationship between them, as shown in Figure 2-11 below.

23
EI
I P P
Spatial Scalability
Enhancement
Layer
(e.g. CIF)
BASE
layer
(e.g. QCIF)
EP EP
key:
Figure 2-11 : Spatial scalability
2.5.2.5. Test Model Rate Control Methods
[6] highlights the existence of a rate control algorithm called TMN-8, which is described in
[27]. Although not an integral part of the H.263+ standard, this algorithm is mentioned here
as it is incorporated within the video encoder software used on this project. TMN-8 controls
the dynamic output bit rate of the encoder according to a specified target bit rate. It achieves
this by adjusting the macroblock quantisation coefficients. Larger coefficients will reduce the
bit rate and smaller coefficients will increase the bit rate.
2.5.3. HiperLAN/2
This section presents a brief overview of the HiperLAN/2 standard, and is largely based on
[15].
2.5.3.1. Overview
HiperLAN/2 is an ETSI standard (see [2], [15] and [26]) defining a high speed radio
communication system with typical data rates between 6 to 54 Mbit/s, operating in the 5 GHz
frequency range. It connects portable devices, referred to as Mobile Terminals (MTs), via a
base station, referred to as an Access Point (AP), with broadband networks based on IP, ATM
or other technologies. HiperLAN/2 supports restricted user mobility, and is capable of
supporting multimedia applications. A HiperLAN/2 network may operate in one of the two
modes listed below. The first mode is more typical for business environments, whereas the
second mode is more appropriate for home networks:
a) “CENTRALISED mode”: where a one or more fixed location APs are configured to
provide a cellular access network to a number MTs operating within the coverage area.
All traffic passes through an AP either to an external core network or to another MT
within the coverage area.
b) “DIRECT mode”: where MTs communicate directly with one another in an ad-hoc
manner, without routing data via an AP. One of the MTs may act as a Central Controller
(CC), which can provide connection to an external core network.
ETSI only standardised the radio access network and some of the convergence layer
functions, which permit connection to a variety of core networks. The defined protocol layers
are shown in Table 2-9 below, and these are described in more detail in each of the following
sections.

24
Table 2-9 : HiperLAN/2 Protocol Layers
Higher Layers
(not part of the standard)
CONVERGENCE LAYER (CL)
DATA LINK CONTROL LAYER
(DLC)
With sub-layers:
 Radio Link Control (RLC)
 Medium Access Control (MAC)
 Error Control (EC)
PHYSICAL LAYER (PHY)
Since these layers, in themselves, do not provide the full functionality necessary to convey
video applications, other functionality in higher protocol layers is required. Table 2-10
presents a generic view of higher layers which would operate above the HiperLAN/2 protocol
stack. The table lists some typical functionality that is pertinent to video coding. This layered
stack, based on the Open Systems Interconnect (OSI) model, is not mandatory for the
development of communication system, however it does provide a useful framework for
discussing where functionality exists within the system.
Table 2-10 : OSI Protocol Layers above HiperLAN/2
Protocol Layers
above HiperLAN/2
Generic Functions within this layer
Application Layer:  Interface to user equipment (cameras/display monitors)
 Determine information formats used in lower layers
Presentation Layer:  Error Recovery
 Video Compression/Coding
 Insertion of synchronisation flags and resynchronisation.
Session Layer:  Set up and tear down of each session.
 Determination of session parameters
(e.g. which layers to use in a layered system)
Transport Layer:  Packet Formation.
 Packet loss detection at receiver.
 Error Control coding.
Network Layer:  Addressing and routing of between physical nodes in the
network.
2.5.3.2. Physical Layer
The physical (PHY) layer provides a data transport service to the Data Link Control (DLC)
layer by mapping and conveying DLC information across the air interface in the appropriate
physical layer format. Prior to providing a brief discussion of the sequence of actions required
to transmit data in the physical layer format, the following feature of the physical layer is
highlighted since it is of prime consideration for this project.

25
2.5.3.2.1. Multiple data rates
The data rate of each HiperLAN/2 connection may be varied between a nominal 6 to 54
Mbit/s, by configuring a specific “PHY mode”. Table 2-11 below lists these modes, showing
the different modulation schemes, coding rates (produced by different puncturing patterns of
the main convolutional code), and the nominal bit rate.
Table 2-11 : HiperLAN/2 PHY modes
Mode Sub-carrier
Modulation scheme
Coding Rate
R
Nominal
(maximum)
Bit Rate (Mbits/s)
1 BPSK 1 / 2 6
2 BPSK 3 / 4 9
3 QPSK 1 / 2 12
4 QPSK 3 / 4 18
5 16QAM 9/16 27
6 16QAM 3 / 4 36
7 64QAM 3 / 4 54
Notes on PHY modes:
 Multiple (DLC) connections may exist simultaneously between a communicating AP and
MT, and each of these connections may have a different PHY mode configured.
 The responsibility to control PHY modes is designated to a function referred to as “link
adaptation”. However, since this is not defined as part of the standard, proprietary
algorithms are required.
 The PHY modes have differing link quality requirements to achieve their nominal bit
rates. Specifically, modes with higher data rates require higher carrier-to-interference
(C/I) ratios to be able to achieve near their nominal data rates. This is shown in the plot of
simulation results provided by M. Butler in Figure 2-12 below.
Figure 2-12 : HiperLAN/2 PHY mode link requirements

26
2.5.3.2.2. Physical Layer Transmit Sequence
In the transmit direction, the sequence of operations performed on data received from the
DLC prior to transmission across the air interface is shown in Figure 2-13 below, and a brief
description of each action is given following this.
1.
scrambling
5.
sub-carrier
modulation
4.
Interleaving
2.
convolutional
coding
data from
DLC layer
8.
RF Modulation and
Transmission
7.
PHY burst
formatting
6.
OFDM
3.
puncturing
Figure 2-13 : HiperLAN/2 Physical Layer Transmit Chain
Transmit Actions:
1. Scrambling: This is performed to avoid long sequences of ones or zeroes
occurring, which can result in undesirable DC bias.
2. Convolutional
encoding:
Channel coding provides error control to improve performance.
Note that the decoding process can apply hard or soft decision
decoding. Soft decision typically improves coding gain by
approximately 2dB, at the expense of increased decoder
complexity.
3. Puncturing The standard convolutional coding above is modified by
omitting some transmitted bits to achieve a higher coding rate.
Different “puncturing patterns” are applied dependant on the
PHY mode selected to achieve the code rates in column 3 of
Table 2-11.
4. Interleaving: Block interleaving is applied as it to improves resilience to
burst errors.
5. Sub-carrier
modulation:
Dependant on the PHY mode selected, the modulation scheme
listed in column 2 of Table 2-11 is applied. The order of these
schemes is directly proportional to the maximum data rate
achievable.
6. Orthogonal Frequency
Division Multiplexing
(OFDM):
Strong justifications for applying this technique are that it
offers good tolerance to multipath distortion [17] and to
narrow–band interferers [18]. It achieves this by spreading the
data over a number of sub-carriers. In HiperLAN/2 this is
achieved by an inverse FFT of 48 data sub-carriers to produce
each OFDM symbol. A guard period is inserted between each
OFDM symbol to combat ISI.
7. PHY Burst
Formatting
The baseband data is formatted into the appropriate PHY burst
format according to the phase within the MAC frame that it
will be transmitted.
8. RF modulation The baseband signal is modulated with the RF carrier and
transmitted over the air interface.

27
HiperLAN/2 will be deployed in a variety of physical environments, and [16] identifies six
channel models specified to represent such environments. These models are often used in
literature ([7] and [16]) as common reference points when quoting performance in
HiperLAN/2 simulations. These models are repeated in Table 2-12 below for reference.
Table 2-12 : HiperLAN/2 Channel Modes
Channel
Model
Identifier
RMS
Delay
spread
[ns]
Characteristic Environment
A 50 Rayleigh Office Near Line of Sight
(NLOS)
B 100 Rayleigh Open space/Office NLOS
C 150 Rayleigh Large Open Space NLOS
D 200 Rician Large Open Space NLOS
E 250 Rayleigh Large Open Space NLOS
2.5.3.3. DLC layer
This layer provides services to higher layers in the following two categories, discussed below.
2.5.3.3.1. Data transport functions
There are two components to data transport: Medium Access Control (MAC) and Error
Correction (EC).
Medium Access Control
In HiperLAN/2, Medium Access Control is managed centrally from an AP, and it’s purpose is
to control the timing of each MT’s access to the air interface, such that minimum conflict for
this shared resource is achieved. MAC is based on a Time Division Duplex/Time Division
Multiple Access (TDD/TDMA) approach with a 2ms frame format, divided into a number of
distinct phases, each dedicated to conveying specific categories of information in both uplink
and downlink directions. An MT is required to synchronise with the MAC frame structure
and monitor information in the broadcast phase of the frame, before it can request use of radio
resources from the AP. An MT requests resources by an indication in the “Random Access
phase” of the MAC frame. Upon detection of this request, the AP will dynamically allocate
timeslots within the downlink, uplink or Direct Link phases of the frame for the MTs to use
for ongoing communications.
Within the DLC layer, data is categorised and passed along two distinct paths, referred to as
the “Control Plane” and “User Plane”. The Control Plane contains protocol signalling
messages, typically used to dynamically establish and tear down connections between the two
communicating end points. The User plane contains actual “user data” from higher layers.
Both categories of data are conveyed in the Downlink, Uplink or Direct Link phases of a
MAC frame, and are contained in one of 2 formats; namely the “long PDU” or “short PDU”
formats. Control plane data is contained in the 9 octet “short PDU”, and User Plane data is
contained in the 54 octet “long PDU”. In the remainder of this report, reference to a
HiperLAN/2 “packet” implies this long PDU structure, which is shown in Figure 2-14 below.

28
PDU type
(2 bits)
Sequence Number
(10 bits)
Payload
(49.5 octets)
CRC
(3 bytes)
 total length: 54 octets 
Figure 2-14 : HiperLAN/2 “Packet” (long PDU) format
Error Correction
This function is applicable to User Plane data only. When activated on selected connections;
it adds error protection on the data. Further details of this facility are covered in section 3.2.7,
as it is of direct relevance to this project.
2.5.3.3.2. Radio Link Control functions
Table 2-13 briefly lists the three main functions with Radio Link control. Further details on
these functions are not presented as they are not directly relevant this project.
Table 2-13 : Radio Link Control Functions
Function Responsibilities of this function
Association
Control
 Registering/Deregistering of MTs with their AP, by
(de)allocation of unique identities (addresses) for these MTs used
in subsequent signalling.
 Configuration of connection capabilities for each MT (for
example, which convergence layer to use, and whether
encryption is activated).
 Security: Authentication and Encryption management.
Radio Resource
Control
 Frequency Selection.
 Power control.
 Handover.
 Transmit Power control.
DLC control  Setup and tear down of DLC connections.
2.5.3.4. Convergence Layer (CL)
Convergence layers adapt the HiperLAN/2 DLC layer to various core networks, and provide
connection management across the interface. There are two categories of convergence layers
defined in the standard, namely:
 Packet-based CL: This integrates HiperLAN/2 with existing packet based
networks such as Ethernet and IP networks.
 Cell-based CL: This integrates HiperLAN/2 with cell-based networks such as
ATM and UMTS networks.

29
Chapter 3 Review of Existing Packet Video Techniques
3.1. Overview
Consistent with the style of contemporary communications standards, H.263+ and
HiperLAN/2 specifications provide enough detail such that systems from different vendors
may inter-operate. Specifications do not typically indicate how to optimise use of various
parameters and options available within the standards. The few occasions when specifications
do provide any such indications may be to clarify mandatory options to achieve inter-working
with another standard, such as illustrated in the convergence layers for HiperLAN/2.
Therefore, the development of techniques to optimise performance of system using these
standards are left to the proprietary endeavours of equipment designers, so as create product
differentiation and competition in the marketplace. This chapter introduces a selection of
these techniques.
This chapter looks at techniques which are generally applicable to video coding (and
independent of the type of transmission channel), as well as focussing on some specific
techniques which are applicable to layered video operating in a wireless packet environment.
To allow further description of these techniques, they are classified under the following
general strategies.
Table 3-1 : General Strategies for Video Coding
Strategy Overview
Algorithmic
Optimisation
(of parameters
and options)
Techniques in this strategy aim to define optimal settings or dynamic
control algorithms for system parameters or options, such that the
combined effect results in optimal performance for a given video
application.
Error
Control
Techniques here aim to reduce the occurrence of errors presented to the
video decoder. By exploiting knowledge of both a) the format of coded
video bitstreams, and b) the differing impact of errors in different
portions of the bitstream; it is possible to provide greater protection for
parts of the data considered more important than others.
Error
Concealment
Techniques here deal with the residual errors that are presented to the
video decoder. Techniques here modify compression algorithms, so that
when errors do occur, these are handled in such a way that:
a) The propagation effects of errors are minimised.
b) Decoding performance degrades more gracefully as conditions in the
transmission degrade.
Techniques are listed in Table 3-2, which provides an indication of which strategies are
incorporated in each technique. A brief description of these techniques is presented in the
following sections, which also contain specific examples taken from existing research. Note
that this list of techniques is not intended to be exhaustive or rigorous.

30
Table 3-2 : Error Protection Techniques Considered
No. Technique Algorithm
Optimisation
Error
Control
Error
Concealment
1. Scaled/Layered parameters
(e.g. number of layers, relative bit rates, modes)
 
2. Rate control.  
3. Lost packet Treatment. 
4. Link adaptation – optimisation of throughput 
5. Intra Replenishment.  
6. Adaptive encoding with feedback.  
7. Prioritisation Packet queuing and dropping.  
8. Automatic Repeat Request (ARQ). 
9. Forward Error Correction (FEC). 
10. Unequal Packet-loss Protection (UPP).  
11. Improved resynchronisation.  
12. Error Resilient Entropy Coding (EREC).  
13. Two-way VLC coding.  
14. Lost Content Prediction and Replacement 
3.2. Techniques Considered
3.2.1. Layered Video and Rate Control
The primary benefit of scaleable video is that it allows the following chain of possibilities:
Defining
layers in
video
bitstream
Prioritisation
to be assigned
to these layers
 Rate Control to be selectively
applied to these priority layers.
 Unequal Packet Loss Protection
(UPP) to be applied to these
priorities.

allows

allows
Scalability was created specifically with the intent to allow video performance to be
optimised depending on both bandwidth availability and decoder capability. A layered
bitstream permits a scalable decoder to select and utilise a subset of the lower layers within
the total bitstream suitable to the decoder’s capabilities, while still producing a usable, albeit
reduced quality video output. Scalable video is less compression efficient than non scalable
video [25], however it offers more flexibility to changing channel bandwidths.
A rate control function may be employed at various nodes in a network to dynamically adapt
the bit rate of the bitstream prior to onward transmission over the subsequent channel.
Typically the rate control function is controlled by feedback indications from the transmission
channel and/or the decoder as well. In a broadcast transmission scenario, with one source
video bitstream distributed to many destination decoders, it is possible for multiple rate
control functions to be employed simultaneously to adapt to the unique bandwidth and
display capabilities of each of the destination decoders. In those cases where rate control
reduces the transmitted bit rate in a congested network, this reduces bandwidth demands on
the network, which may actually serve to reduce congestion in the network. The reduction in

31
bit rate is typically achieved by dropping frames from higher (lower priority) layers until an
acceptable threshold is reached.
While the video coding standards define the syntax for layered bitstreams, settings such the
relative bit rates of each layer, and algorithms to prioritise and protect these layers are left to
the discretion of system designers. [25] is highly relevant, as it contains the following
findings on layered H.263+ video settings:
1. Coding efficiency decreases as the number of layers increases.
2. Layered video can only be decoded at the bit rate at which is it was encoded.
3. Based on point 2, it is therefore necessary to know the most suitable bit rates to encode
the base and enhancement layers at prior to encoding. However, this constraint can
eliminated if an adaptive algorithm using feedback from the transmission channel and/or
the decoder is applied to dynamically adjust settings at the encoder.
4. Informal observation trials feedback indicates that:
a) Repeated noticeable alterations to quality (e.g. by repeated dropping and picking up of
enhancement layers) is displeasing to viewers.
b) A subset of users desire the ability to have some control of the scalability/quality
settings themselves. This necessitates a feedback channel (as suggested in point 3
above).
5. SNR scalability:
a) The frame rate is determined by the bit rate of the base layer. Combining a low base
rate with a high enhancement rate is not considered appropriate (as the frame rate will
not increase). In this case it is recommended to apply other scaling modes (i.e.
temporal or spatial) at the encoder.
b) There is relatively less performance benefit when going from two to three layer SNR
scalability.
c) For low base layer bit rates, it is recommended to apply SNR scalability (as opposed
to temporal or spatial) at the first enhancement layer - as this provides the largest
PSNR performance improvement at these low bit rates.
6. Temporal scalability:
a) Spatial quality of base and enhancement layers is determined by the bit rate of the
base layer.
b) When the base rate is very low (9.6kbps or less), it is not recommended to use
temporal scalability mode.
c) Temporal scalability should only be applied when the base layer bit rate is sufficient
(at least 48 kbps) to produce acceptable quality P-frames.
d) As the enhancement layer bit rate increases, the frame rate will increase - but only up
to the maximum frame rate of the original sequence.
e) Bandwidth for the temporal enhancement layer should not be increased further once
the maximum frame rate has been reached.
f) Once the bit rate between base and enhancement layers exceeds 100 kbps, it is
recommended to apply an additional layer of another scaling mode.
7. Spatial scalability:
a) This only brings noticeable benefit in PSNR performance when there is high
bandwidth available in the enhancement layer.
b) Combined base and enhancement layer bit rates of less than 100 kbps result in
unusable quality.
c) Frame rate is largely determined by the base layer bit rate.

32
d) It is not recommended to interpolate a QCIF resolution frame for display on a CIF size
display, as this actually makes artefacts appear more pronounced to a viewer.
3.2.2. Packet sequence numbering and Corrupt/Lost Packet Treatment
HiperLAN/2 uses a relatively short, fixed length packet to convey coded video data. It is
therefore typical that data for one frame will be spread over numerous sequential packets. As
packets are sequence numbered and re-ordered prior to presentation to the video decoder, it is
possible to detect lost packets within a frame. Potential treatments by protocol layers
following detection of corrupt or lost packets include those listed in Table 3-3 below.
Table 3-3 : Lost/Corrupt Packet and Frame Treatments Considered
Treatment Description
Packet
Treatment
skip packet simply omit the errored packet(s)
Packet
replacement:
 zero fill packet
 ones fill packet
Replace the contents of lost or errored packet(s)
with set patterns and forward these to the decoder.
Set patterns are chosen with intention that the
decoder notices coding violations early within this
packet, and attempts resynchronisation
immediately. The desired outcome is that the
decoder regains synchronisation earlier than it
would have done if the packet was skipped.
The two patterns considered are all zeroes and an
all ones pattern (this all ones is actually modified
to avoid misinterpretation as an End of Sequence
(EOS) character).
Frame
Treatment
skip frame Skip the entire frame which contains errored
packets.
[9] applied this technique to enhancement layer
frames only. The intended benefit of this is that if
the decoder loses synchronisation towards the end
of an enhancement layer frame, the
resynchronisation attempt may actually
misinterpret the start of the next (base layer)
frame, resulting in corruption within that frame
too. Skipping the errored enhancement frame
therefore ensures that the next base frame is not
impacted.
abandon remaining
packets to end of
frame
Instead of skipping the entire frame, this only
omits the portion from the (first) corrupt packet
onwards. This can benefit the decoder where
significant amount of valid data is received at the
start of the frame is received. However this option
suffers the potential to impact the data at the start
of the next (otherwise valid) frame while re-
synchronising.

MSC-Thesis-Wireless-Video-over-HIPERLAN2-JustinJohnstone

MSC-Thesis-Wireless-Video-over-HIPERLAN2-JustinJohnstone

Recommended

Recommended

More Related Content

Viewers also liked

Viewers also liked (15)

Similar to MSC-Thesis-Wireless-Video-over-HIPERLAN2-JustinJohnstone

Similar to MSC-Thesis-Wireless-Video-over-HIPERLAN2-JustinJohnstone (20)

MSC-Thesis-Wireless-Video-over-HIPERLAN2-JustinJohnstone